Unsupervised image segmentation using Fuzzy Hidden Markov Chain with bi-dimensional data

This work deals with a method based on a fuzzy hidden Markov chain with bi-dimensional data for unsupervised segmentation of fuzzy images. Often, to apply the hidden Markov chain model on image segmentation we use a curve like Hilbert-Peano scan to transform a 2 D image to 1 D. The main characteristic of our approach is using it as an alternative of Hilbert-Peano scan and to treat the coexistence of hard and fuzzy pixels in the same image. We use the EM algorithm in bi-dimensional version (EM2D) for estimation of parameters in this model. We test the EM2D algorithm with the EM in one-dimensional case (EMID) with Hilbert-Peano scan by examining some criteria as segmentation quality and the error rate. The segmentation results demonstrate the quality of our model.